Researchers at Samsung's AI Center in Moscow have developed a new system that can transform facial images into videos with speech expressions.
According with a study, the system creates realistic virtual talking faces through an application that processes the target person's face from a still photo. The AI application can control the way a person moves to display realistic results.
"This capability will have practical applications in video surveillance, in some video conferencing, in multiplayer games as well as in the special effects industry," Samsung said.
Deepfake technology is not new, but Samsung's new system does not use 3D models and requires only one face photo to create a new face model that moves and speaks.
If the application is able to use 32 images to create a model, it will be able to "achieve [a] perfect realism and a very good personalization", according to the company.
The ability to create a "deepfake" video with a limited number of downloads is due to the fact that the system has a large database of videos of talking people, according to Samsung. Using this database, combined with images of target persons, the system is able to create several realistic models of the same person.
So-called "deepfake" videos are now a major concern for U.S. lawmakers, who are worried that AI that manages human videos could create scenes that say things that have not been said, depending on the creator's mood.
Imagine what applications the new AI technology can have, in the case of fake news.
- Acer notebooks Nitro 5 & Swift 3 with AMD Ryzen Mobile 2nd generation
- Laptop vs. desktop. Advantages, disadvantages, and what to buy
- How to Upgrade Windows 10 with USB, DVD or Local Media